ChatGPT ads can target people by uploaded phone or email

Anyone with a list of email addresses, phone numbers or mobile device identifiers can upload it to OpenAI and show ads to exactly those people inside ChatGPT. 404 Media found this in its own tests of ChatGPT’s ad system, published on October 16. Joseph Cox, the reporter who ran the test, says the setup took him five minutes.
Cox’s test ad was about ChatGPT itself. It told people that ChatGPT is not private and that OpenAI hires human contractors to read users’ prompts. He showed it to people asking ChatGPT about privacy, about AI and about whether their chats are private. That is the second way to target ads in ChatGPT, and the one the outlet calls new. Targeting ads at an uploaded list of emails or phone numbers is somewhat standard practice across the ad industry, 404 Media says. Targeting based on what people type into the chatbot is not, and users may assume those prompts are private. The outlet has previously reported that humans may review them.
The door Cox walked through has been open to small buyers for months. OpenAI announced an ads test on January 16, and on February 9 ads went live as “Sponsored” blocks below answers. They were shown to logged-in US adults on the Free and Go tiers, with health, mental health and politics excluded at launch. On May 5 OpenAI opened its self-serve Ads Manager to all US businesses, dropped the minimum spend and added cost-per-click bidding. With no minimum spend, the word “anyone” in Cox’s account is close to literal.
The audience is large. 404 Media puts ChatGPT at one billion monthly active users, and says many of them may not realize that some ads reach them because an advertiser uploaded their personal data to OpenAI.
Ads appear only on the free tiers. Plus, Pro and Business plans have never shown them, and free users can turn ads off in exchange for fewer messages a day.
Up next
ClaudeClaude Codeanalysis
Context-tax prices what Claude Code sends before you type
Measured on one real machine: a single built-in Claude Code tool weighed 11,402 tokens of a 47,700-token first request. The context-tax CLI reads your transcripts to price yours. Is your Claude Code context trimmed by hand right now or left at defaults?
Related stories
- OpenAI’s ad pixel fires before consent on 224 of 341 shops
- OpenAI’s ChatGPT ads hit a $1B run rate in under 200 days
- Security researchers: stop blocking defenders
- GPT-6.2 Astra costs 20% less than GPT-6 Astra in the API
- Defense says no evidence OpenAI bugging suspect is a spy
- NATO’s cyber agency used up to 300B OpenAI tokens a month
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

